Video surveillance system for fall detection at home

Abstract

Human fall detection is an area of increasing interest in in today’s world, especially

due to the aging population, the disabled and young children. In order to avoid the invasiveness that most methods produce, silhouette areas from videos are instead used for this purpose. This project uses a multi camera fall dataset to simulate falls. Matlab is used as a platform to achieve the purpose of this project. Silhouette areas are obtained with the use of two related methods and two different types of feature vectors are identified. Two support vector machines are trained with the normalised data samples obtained from the features, and are used to classify falls based on the characteristics identified. The effectiveness of the system is determined with calculations to measure its success, which include the accuracy the ability to identify falls with few false alarms. A few different kernel parameters are tested, after which

the kernel and parameters for each feature and the best feature are chosen. A conclusion is also made at the end along with recommendations to improve on the outcome.